Rheumatology Advance Access originally published online on August 13, 2007
Rheumatology 2007 46(9):1388; doi:10.1093/rheumatology/kem161
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
EDITORIALS |
Problems with causality
Complementary Medicine, Peninsula Medical School, Universities of Exeter & Plymouth, 25 Victoria Park Road, Exeter EX2 4NT UK
Correspondence to: Edzard.Ernst{at}pms.ac.uk
In this issue, Williamson et al. [1] present an interesting pragmatic trial of acupuncture vs supervised physiotherapy vs written instructions for home exercise (the control group) in patients with knee osteoarthritis booked for joint replacements. The results suggest a marginal benefit for acupuncture compared with the other treatments. We tend to attribute such differences to the respective intervention. On closer inspection, however, some doubts seem justified.
The perceived therapeutic effect observed in a group of patients of a randomized controlled trial (RCT) can be caused by a host of factors [2]. They include:
- the natural course of the condition,
- undeclared concomitant treatments,
- regression towards the mean.
The purpose of the RCT design is to eliminate these factors; if they influence the outcome in all groups in a similar fashion they do not affect any inter-group comparisons. But there are other factors which may not always be distributed equally between groups and could thus impact on inter-group differences of RCTs. They include:
- the placebo effect,
- social desirability,
- therapist–patient interactions,
- disappointment of being allocated to the control group,
- and, of course, the specific therapeutic effects of the treatment(s) under scrutiny.
Contrary to a common assumption, the placebo effect is not a constant. If today acupuncture is ‘in’, it may generate higher expectations, and hence a more pronounced placebo response than physiotherapeutic exercise. Once the novelty factor wears off, its placebo-effect could decrease.
Social desirability describes the phenomenon that patients tend to be nice to us when we are nice to them. Thus, they may say that their symptoms have improved, if in fact they have not. In the trial by Williamson et al. [1], this could have contributed to an apparent superiority of acupuncture and physiotherapy compared with control.
The therapist–patient relationship is a highly therapeutic element in most clinical settings and a significant contributor to clinical outcomes [3]. In the current study [1], two patient groups may have benefited from this factor while the control group did not.
In the Williamson study [1], the control group received an exercise and advise leaflet. It is not difficult to imagine that some of these patients were disappointed after not being allocated to one of the treatment groups. It is conceivable, therefore, that this disappointment affected the study results.
Finally, there is the specific effect of each treatment tested in an RCT. In principle, the specific effect can be positive, negative or non-existent. Given the multitude of variables contributing to any inter-group difference in RCTs, it is difficult to be sure that, in the current study [1], the specific effect of acupuncture was positive. In fact, it is theoretically conceivable that the specific effect of acupuncture was negative, i.e. made symptoms not better but worse. Due to the positive influence of the other contributing factors, the outcome in comparison to the control group could be positive even if the specific effect of acupuncture is negative (Fig. 1).
|
Why does all this matter? It matters, I think, in several ways. Firstly, it reminds us how cautious we often have to be when making causal inferences based on the results of pragmatic RCTs. Second, it emphasizes the importance about minimizing confounders in RCTs. The majority of the confounding variables can be eliminated by using blinding and credible placebos in RCTs. Patient blinding in acupuncture trials is now possible with the non-penetrating sham-needles that recently have become available [4]. Third, it highlights the difference between pragmatic (effectiveness) and fastidious (efficacy) RCTs. If healthcare decisions are based only on pragmatic trials, there is a risk adopting interventions with no or even negative specific effects. Whenever possible, therefore, we need to have both pragmatic and fastidious RCTs—and ideally the latter should precede the former.
The author has declared no conflicts of interest.
References
- Williamson L, Wyatt M, Yein K, Melton J. Severe knee osteoarthritis: a randomised controlled trial of acupuncture, physiotherapy (supervised exercise) and standard management for patients awaiting knee replacement. Rheumatology (2007) June 29.
- Ernst E. A schematic analysis of placebo effects in clinical trials. Perfusion (2007) 20:1–3.[CrossRef]
- Di Blasi Z, Harkness E, Ernst E, Georgiou A, Kleijnen J. Influence of context effects on health outcomes: a systematic review. Lancet (2001) 357:757–62.[CrossRef][ISI][Medline]
- Ernst E. Acupuncture - a critical analysis. J Intern Med (2006) 259:125–37.[CrossRef][ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||